基于拉曼光谱的已知混合物组分定量分析方法

Quantitative Analysis Method for Mixture With Known Components Based on Raman Spectroscopy

利用拉曼光谱进行混合物组分定量分析一直是分析化学领域的一大难题。针对现有的基于机器学习(如支持向量回归机、偏最小二乘)的混合物定量分析方法存在的训练样本难以获得、模型推广性能差的问题,提出了一种基于拉曼光谱谱峰强度最小二乘拟合的已知混合物组分直接定量分析方法。该方法首先采集已知混合物及其各组分的拉曼光谱,利用连续小波变换和惩罚最小二乘法相结合的方法对采集的拉曼光谱进行去噪、基线校正等预处理。通过斜率比较法将预处理后的拉曼光谱分为多个光谱子区间,将各子区间的拉曼光谱看作是多个Voigt函数的线性叠加,并利用levenberg-marquardt-fletcher(LMF)算法求解获得各谱峰的位置、强度、半高宽等表达系数。根据各组分参考光谱的谱峰位置,确定各组分对混合物光谱中每个谱峰的贡献度。依据朗伯-比尔定律中拉曼光谱的谱峰峰强与其所对应的浓度的正比关系,建立超定方程;最后利用最小二乘法拟合该超定方程得到各组分对应的系数,从而获得各组分的体积浓度。利用乙醇、乙腈、丙酮、环己烷、二丙酮醇、丙二酸二乙酯六种组分配置了10种三元混合物(每种三元混合物9个体积浓度比),采集了90组混合物及6种组分的拉曼光谱数据。在混合物及其组分参考光谱测量条件(功率和积分时间)相同情况下,所有组分的相关系数(r)均在0.96以上,均方根误差(RMSE)小于6%,剩余预测偏差(RPD)均大于2.5;在混合物及其组分参考光谱在不同测量条件下,各组分的r均大于0.93,最大RMSE为7.94%, RPD均大于2.0,证明了算法具有良好的准确性和鲁棒性。所提出的方法能够实现对三组分混合物的快速、准确的直接定量分析,为混合物的定量分析提供了一种有效的途径。

Quantitative analysis of mixture components by Raman spectroscopy is a difficult problem in analytical chemistry.The existing quantitative analysis methods of the mixture based on machine learning(such as support vector regression,partial least squares)have the problems of difficult to obtain training samples and poor generalization performance of models.A direct quantitative analysis method for mixture with known components based on Raman peak intensity and the least square fitting algorithm was proposed in this study.Firstly,the Raman spectra of the mixture and its components were collected respectively,and the noise reduction and baseline correctionforthe Raman spectra were conducted by using the combination of continuous wavelet transform and penalized least square method.Secondly,the preprocessed spectra were divided into several spectral subintervals by slope comparison method,each subinterval was regarded as the linear superposition of several Voigt functions,and the positions,intensities and half-widths of the peaks were obtained by Levenberg-Marquardt-Fletcher(LMF)algorithm.Thirdly,the contribution value of each component to the spectral peak of the mixture was determined based on the peak position.Finally,the over determined equation was established based on lambert-beer law thatthe peak intensity of mixture is proportional to its concentration of components,the coefficients corresponding to each component were obtained by fitting the equation with the least square method,so that the volume concentration of each component was obtained.In this study,ten kinds of ternary mixtures(9 volume concentration ratios of each ternary mixture)were prepared with 6 components,including ethanol,acetonitrile,acetone,cyclohexane,diacetone alcohol and diethyl malonate,and Raman spectral data of 90 mixtures and 6 components were collected.If the spectra of mixtures and their components were acquired at the same measurement conditions(power and integral time),the obtained correlation coefficient(r)foreach componentwas above 0.96,the root means square error(RMSE)was less than 6%,and the residual prediction deviation(RPD)was greater than 2.5.If the spectra of mixtures and their components were acquired at the different measurement conditions,the correlation coefficient(r)was above 0.93,the RMSE was less than 7.94%,and the RPD was greater than 2.0,this proved that the algorithm has good accuracy and robustness for quantitative analysis of mixture based on Raman spectroscopy.The proposed method can achieve a arapid and accurate direct quantitative analysis of ternary mixtures,which provides an effective way for the quantitative analysis of mixtures.